JP3489704B2 - Method and decoder for decoding encoded audio signal, and method and encoder for encoding audio signal - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coding system, and more particularly to parameter quantization in a speech coding system.

[Prior art]

Speech coding systems perform the function of providing a system word receiver with a codeword representation of a speech signal for communication over a channel or network. Each system receiver reconstructs a speech signal from the received codeword. The amount of codeword information communicated by the system within a given time interval defines the bandwidth of the system and affects the quality of speech received by the system receiver.

The purpose of a speech coding system is to determine the quality and bandwidth of speech given the secondary conditions such as input signal quality, channel quality, bandwidth limitation, and cost. It is about providing the best trade-off relationship between them. Audio signals are represented by a combination of parameters that are quantized for transmission. Perhaps in the design of a speech coder, the search for a good combination of parameters to describe the speech signal is of paramount importance. With a good combination of parameters, both perceptually accurate reconstruction of the speech signal requires only the bandwidth of the slow system. In addition, a desirable property for a combination of parameters is that the parameters are independent. When the parameters are independent, the quantizer can be designed independently so that even incorrectly received information will have less impact on the quality of the reconstructed speech signal. Let's do it. The bandwidth required for each parameter is about the accuracy with which the trajectory of the rate at which each parameter changes and the value of the parameter must be described in order to obtain a reconstructed speech with the required quality. Is a function of.

The output of a voice signal is desirable as one parameter of a combination of coding parameters. The other parameters are easily made independent of the signal output. In addition, the signal output is an indication of the physical characteristics of the audio signal and facilitates defining design criteria for the quantizer. The signal output shall be defined as the energy of the signal per sample, averaged over one pitch period for the quasi-periodic speech part and over some predetermined time interval for the aperiodic speech part. Is possible. The time intervals for aperiodic parts should be short enough to be perceptually significant (important).
(5 ms or less is effective.) By using such a definition, the output of the voice signal becomes a smooth function between the continuous vowels, and the onset and plosive sound are clearly shown.

Estimation of high-resolution signal output (estimation)
Cannot be obtained with a fixed or large window size. The large window size for estimation leads to a low temporal resolution for the estimated signal power. As a result, speech reconstructed with a low rate encoder using such an approach will generally lack sharpness. On the other hand, short, fixed windows lead to instabilities in the signal output. Because of this, CELP (Code-E
xcited-Linear-Predictive,
An encoder using a short fixed window, such as a Code-Excited LPC method, generally does not use the signal output as an explicit parameter.
(For example, BSAtal, "High-Quality Speech at L
ow Bit Rates: Multi-Pulse and Stochastically Excite
dLinear Predictive Coders, "Proc.Int.Conf.Acoust.Sp
eech Sign. Process., Tokyo, pp. 1681-1684, 1986, etc. )

With the increasing demand for coding efficiency,
It is expected that more encoders will use the signal output as an explicit parameter for independent encoding. Recently, a coding process has been introduced which describes a speech signal in terms of its characteristic waveform and is sampled at a high rate (approximately 500 Hz). (For example, WBKleijin and J. Haagen, "Transformati
on and Decompositionof the Speech Signal for Codin
g, "IEEE Signal Processing Letters, Vol.1, September
1994, pp. 136-138. In these encoders based on so-called waveform interpolation, the estimation (estimation) window of the signal output is one pitch period.
These new waveform interpolation encoders (for speech)
The analysis is used to estimate the signal output very accurately with a high time resolution. The signal output is encoded independently.

[Problems to be Solved by the Invention]

In conventional coding techniques that use the signal output as an explicit parameter, the signal output is transmitted at a fairly low rate (low bit rate). Linear interpolation over time intervals that are updated over time is then used to reconstruct the contour of the signal output. (Such interpolation is often applied to the log representation of the output.) (For example, TETremain, "The Governm
ent Standard Linear Predictive Coding Algorithm, "S
See peech Technology, pp40-49, April 1982. ) Making a more detailed description of the output contours will improve the quality of the reconstructed signal. However, the problem remains to transmit only the perceptually important details of the contour of the signal output, and low bit rates can still be used.

[Means for Solving the Problems]

The present invention provides a method and apparatus that allows transmission of perceptually significant features of speech coding parameters at low bit rates. For example, the audio coding parameters may include audio signal output. The parameters are processed on a block-by-block basis. The parameter value at the block boundary is transmitted by a conventional method such as a means for differential quantization. Therefore, in the present invention, the contour shape of the reconstructed parameter within the block boundary is based on the classification. The classification relies on the perceptually important features of the contours of the parameters within the block. The classification can be done at the transmitter end of the encoder (eg, using the contour of the original parameter with high temporal resolution or other similar possible audio parameters) or at the receiver end of the encoder (eg, transmitter Parameter values, as well as other similar possible transmitted voice parameters). Based on the result of the classification, as well as the values of the parameters at the boundaries of the blocks, the parameter contours (within the blocks) are selected from the inventory of possible parameter contours.

[0009]

DETAILED DESCRIPTION OF THE INVENTION

[Introduction] The goal of speech coding is desirable within the trade-off between reconstructed speech quality and required bandwidth, subject to channel quality, hardware, and delay constraints. The thing is to get things.
In general, a model is used for a voice signal, and a locus of a model parameter (which can be a vector) as a function of time is transmitted with a certain accuracy. (In the simplest model, the model parameter is the speech signal itself.) In the digital speech coder, the locus of the model parameter is described as an array of scalar quantity or vector quantity samples. These parameters may be transmitted at a low speed (low bit transmission rate), and the locus is reconstructed by interpolation between the updated points. Alternatively, a predictor (which may be a linear predictor) is used to predict the parameters from the previously reconstructed samples, and the difference between the actual and predicted values. Only (residual) is transmitted. In yet another process, a high temporal resolution description of the parameter trajectory may be divided into sequential blocks. Such sequential blocks are quantized vectors for transmission. Some encoders combine vector quantization and prediction.

In an exemplary embodiment of the invention, the parameter trajectory (which may be a vector) is:
It is transmitted by a method that increases the methods of interpolation, prediction, and vector quantization described above. The parameters are transmitted on a block-by-block basis, and each block contains a plurality of parameter samples at the analysis side. The parameter signal is low-pass filtered and down-sampled. This downsampled parameter array is transmitted according to conventional means. (For example, in the exemplary embodiment described in the next section, this conventional transmission uses a differential quantizer.) At the receiver, the parameter array is reconstructed by the speech model. It must be upsampled to the rate needed to be done. Obviously, signal features are lost when band limiting or linear interpolation is used for upsampling. In an exemplary embodiment of the invention, classification is used to identify perceptually important features of the parameter trajectory, otherwise the reconstructed parameter is based solely on interpolation. In the array, there will be no perceptually significant features of such parameter trajectories. According to the result of such classification, one trajectory is selected from the inventory of trajectories in order to construct the trajectory of the parameters between the samples at the boundary of the block. In addition, this inventory
Is adapted to the parameter values at the block boundaries. The exemplary methods described herein do not necessarily require the transmission of additional information. That is, the classification is performed at the receiver end of the encoder using only the transmitted and downsampled array of parameters.

Exemplary Embodiment In the exemplary embodiment shown here, the process described above is applied specifically to audio output. A stepped contour audio signal sounds significantly different than a smooth contoured audio signal. Smooth contours are typically found in sustained vocal sounds, while stepped contours are common in pronouncing sounds. A simple classification scheme, which uses an array of transmitted and down-sampled audio outputs, makes it possible to reliably identify stepped audio signal contours. There, stepped contours are used for the array of reconstructed signal outputs. Experiments have shown that the exact position of the steps in the signal of the speech output has only a slight importance to the perceived speech quality.

The classification performed at the transmitter end of the encoder can be used to identify features in the contour of energy between samples, such as plosives. Also, the exact location of the reconstructed plosive is
It has negligible perceptual importance. Thus, whenever a plosive is identified at the transmitting end, a simple bulge in the audio output signal will be added to the center of the block.

FIG. 1 illustrates the transmit side portion of an exemplary embodiment of the present invention which performs signal output extraction in a waveform interpolation encoder. The original speech signal is first processed in a coding unit (encoding unit) 101. In the waveform interpolation encoder, this encoding unit (encoding unit) extracts a characteristic waveform. These characteristic waveforms correspond to one pitch period during which the voice is being produced. According to known methods, a speech signal is represented by a characteristic array of waveforms (defined by the difference part of a linear prediction), a trajectory in pitch periods, and a time-varying linear prediction coefficient. Such technology is, for example, US Patent application Ser. No. 08 / 179,831 filed together with the assignee of the present invention.
(US Patent Application No. 08 / 179,831),
"Method and Appratus For Prototype" by WB Kleijin
Waveform Speech Coding ",
By reference, such techniques are incorporated, as well shown fully herein. (In addition,
For example, WBKleijin, "Encoding Speech Using Prototyp
e Waveforms, "IEEE Trans.Speech and Audio Processin
g, Vol. 1, No. 4, pp. 386-399, 1993 and WBKleijin and J.
Haagen, "Transformation and Decomposition of the Spe
ech Signal for Coding, "IEEE Signal ProcessingLette
rs, Vol.1, September 1994, pp.136-138. )

The characteristic waveform description is often in the form of a finite Fourier sequence. The characteristic waveform is described in the remaining (difference) part. This is because this makes extraction and quantization easier.
The sampling (extraction) rate of the characteristic waveform is effectively tuned to approximately 500 Hz. In this figure, as in the following drawings, it is considered that the locus in the pitch period and the linear prediction coefficient can be used for any processing unit (processing unit) that requires these parameters. Both the trajectory in pitch period and the linear prediction coefficient are defined and interpolated according to conventional methods.

The unquantized portion of the characteristic waveform (indicated as an unquantized intermediate signal in FIG. 1) is supplied to the output extractor 102. In the output extractor 102, the remaining (difference) portion of the characteristic waveform is first converted into the speech portion of the characteristic waveform by means of cyclic convolution using a linear prediction synthesis filter. (This kind of convolution can be performed, for example, by WBKleijin, "Encoding Speech Using Pro
totype Waveforms, "IEEE Trans.Speech and AudioProce
It can be directly executed based on the Fourier sequence by means of the equation (19) of ssing, Vol. 1, No. 4, pp. 386-399, 1993. The signal output in the audio part is used because transmission errors in the linear prediction coefficients do not affect the output of the audio signal.

Therefore, the output extractor 102 calculates the output of a characteristic waveform for each voice sample. The output is normalized on a sample-by-sample basis so that the signal output is independent of pitch period. This facilitates quantization and makes the output less susceptible to errors in the communication path that affect the pitch period. Finally, the output extractor 102 takes the resulting output of the audio portion as the logarithm (display) of the output of the audio portion.
Convert to. For example, the widely known decibel (d
The log scale (display) of B) could be used for such purpose. (Using the logarithm (display) of the signal output rather than the linear signal output is more motivated by the human perceptual characteristics.
It is possible to handle signal outputs that vary over orders of magnitude. ) Such a signal, sampled at the same rate as the characteristic waveform, is fed to a plosive detector 105, a low pass filter 106, a normalizer 103. A normalizer (normalizer) 103 uses the extracted signal output to create a normalized characteristic waveform. Such a normalized characteristic waveform may be further encoded in the encoding unit (encoding unit) 104, and the signal output may be used as additional information.

To avoid aliasing (the same signal is overlaid), the low pass filter 106 removes frequencies above half the sampling frequency for the output signal of the downsampler 107. 2.4 Kb /
For the s encoder, the sampling frequency after downsampling is matched to 100 Hz (corresponding to downsampling with a factor of 5 in the example given here). Is effective.

Output encoder (output encoder) 108
Encodes the downsampled log representation of the output array. This is more effectively processed using a differential quantizer. Here, assume that the output of the log display at the sampling time n is x (n). Then, in order to quantize the difference signal e (n), a quantizer with a simple scalar quantity is used, that is, e (n) = x (n) −α ^* x (n−1) (1) Can be expressed as Let Q (e (n)) denote the quantized value of e (n). Then the reconstructed lo
The output in g display is X (n) = Q (e (n)) + α ^* x (n−1) (2). For α less than or equal to 1, equation (2) shows the well known leaky integrator. leaky integral
The function of r (leakage integrator) is to reduce the sensitivity to errors in the channel. α = 0.
A value of 8 is effectively used.

The plosive detector 105 is an array of unprocessed log output and low-pass filtered.
Use the output array of og display. Each time interval between samples of the array of downsampled log representations (eg, 10 Hz based on a downsampled sampling rate of 100 Hz).
The output of the plosive detector (for example, ms) is two decisions. That is, 1 means that a plosive sound is detected, while 0 means that a plosive sound is not detected.

The operation of plosive detector 105 is shown in FIG. The peak clearance detector 304 is
A determination is made as to whether the sample of the og display output minus the array of the low display filtered log display output of the same sample is greater than a given threshold. (For example, such a threshold is effectively matched to 16 dB for log representations of signal output.) If this is the case, the output of peak clearance detector 304 is 1, otherwise. The output is zero.

The operation of the hat hanger 301 is illustrated in FIGS. 5 and 6. Conceptually, a hat-shaped curve is, so to speak, hung from the output signal samples here. That is, the top of the portion corresponding to the hat is leveled with the top of the sample here. The output of the hat clearance detector 303 will be 1 if the sample of the portion covered by the shape of the hat fits below the hat and its periphery. For example, FIG. 5 shows that the hat does not avoid collision with an adjacent sample. Because of this, the output of the hat clearance detector 303 is zero. On the other hand, FIG. 6 shows a situation in which the hat avoids collision with an adjacent sample. Because of this, the output of the hat clearance detector 303 is 1. The characteristics of the hat are stored in the hat keeper 302. The shape of the hat can be changed within the detection interval, and the height of the peripheral portion can be different between the left side and the right side.
For example, if the hat is symmetrical, the width of the top of the hat and the width of the perimeter may be effectively adjusted to 5 ms respectively, and the distance of the perimeter to the top may be For the contour describing the log representation of the output, it may be useful to be tuned to 12 dB. For example, it will be recognized by those skilled in the art that the hat clearance detector 303 can be supplemented with a sample memory and processor for testing sample levels and comparing these levels to a given threshold. Ah

An arithmetic unit 305 having a logical AND function
Combines the output from the peak clearance detector 304 and the output from the hat clearance detector 303.
If any one of these two outputs is 0, the output of the arithmetic unit 305 having a logical AND function becomes 0. The logically ORed downsampler 306 has one output for each time interval, an array of downsampled log representation outputs. (Ie, output of downsampler 107) For example, for the example case described previously, this would be one output every 10 ms. If logically, the down sampler 3 with the function of OR
If the input to 06 is non-zero at any time within such a time interval, then the output of the downsampler 306, which is logically an OR function, is set to one. And this shows that the plosive sound was detected. If the input is zero at any time within the time interval, then the output of the logically ORed downsampler 306 is zeroed. This indicates that no plosive sound was detected.

FIG. 2 shows the receiving part of an exemplary embodiment of the invention corresponding to the transmitting part shown in FIG. The decoder unit 201 reconstructs the characteristic waveform. Decoder unit 2
Some of the operations performed in 01 do not correspond to the operations performed at the transmitter. For example, in order to emphasize the spectral shape of the output signal, the spectrum before the shape is shaped may be added to the characteristic waveform. This means that the characteristic waveform forming the output of the decoder unit 201 is generally not guaranteed to have a normalized output. For this reason, these outputs must be evaluated before scaling (converting) the quantized characteristic waveform. This is done by the output extractor 202, which functions in a manner similar to the output extractor 102. The output is also evaluated in the audio part.

The scale factor processor 206 determines the appropriate scale factor to be applied to the characteristic waveform generated by the decoder unit 201. For each characteristic waveform, the input to the scale factor processor 206 is the log-represented output value reconstructed from the transmitted information, prior to scaling (conversion). It is a quantized characteristic waveform. The output values in log representation are converted to linear output values and divided by the output of the unscaled quantized characteristic waveform. Such division will produce an appropriate scale factor for the unscaled quantized characteristic waveform. The resulting scale factor is used in amplifier 207, which has as output its appropriately scaled and quantized characteristic waveform. This characteristic waveform is the input of a decoder (decoder) unit 203, which provides a description of the array of characteristic waveforms (with the assistance of loci in pitch periods and linear prediction coefficients). ) Convert to reconstructed audio signal. Well-known methods used in the decoder unit 203 are described in US Patent application Ser.N.
08.179,831 (US Patent Application No. 08/179,
831).

Here, the reconstruction of the array of outputs in the log display will be explained. Output decoder (power decoder) 2
04 is the downsampled and quantized log
The display output array is reconstructed based on equation (2) above. Output (power) envelope processor 205
Converts this downsampled array to an array of unsampled log representations of the output. The operation of the output (power) envelope processor 205 is illustrated in detail in FIG. First, the case where the information about the plosive is 0 (indicating that there is no plosive) would be considered. Output step evaluator (power step evaluator) 40
1 is the current lo of the downsampled array
The difference is determined by subtracting the output value from the previous log display of the downsampled array from the output value from the g display. The upsampler 402 upsamples the array of outputs in log display according to the upsampling process. Above all, upsampler 40
The upsampling process performed by 2 is selected based on comparing the difference between successive samples (determined by the output step evaluator (power step evaluator) 401) with a threshold value. For example, the threshold is 12 dB in the log display of the audio output and 100 Hz.
It may be useful to choose a sampling rate of The linear interpolation between the updated points is performed by the upsampler 402 if the difference between consecutive samples is less than the threshold. This is the case for most time intervals and is illustrated in FIG. FIG. 7 shows two sample values with a thick line for the array of the down-sampled output in the log display. The sample between these two sample values is obtained by linear interpolation.

A larger increase in the signal output, such that the difference between consecutive samples exceeds the threshold, occurs mainly at the beginning of a sharply pronounced note. lo
The linear interpolation of the g-display output is not a good model for producing such a sound. Therefore, in such a case, the upsampler 402 utilizes a stepped contour. In particular, whenever the difference between consecutive samples exceeds the threshold, the output value of the left log representation (ie, the previous sample) is used until the midpoint of the time interval and the right log representation is reached. The output value of (i.e., the current sample) is used for the rest of the time interval. For this example,
This is illustrated in FIG. Generally, it is considered that the step is not located at the same (temporal) moment as the start of the original signal. However, the exact position of the steps in the output contour is less important to human perception than the steps are contained in the time interval rather than the smooth contour.

A perceptual effect of the stepped output contour may be to make the reconstructed audio signal noticeably sharper (more crisp). However, the indiscriminate use of stepped output contours results in a significant loss of output signal quality. Limiting the use of step-by-step contours for cases where the signal output is changing rapidly has resulted in an improvement over using consistently consistent use for linearly interpolated contours. You will get the quality of the voice. Moreover, the use of step-by-step contours does not have a significant effect on the reconstructed speech in the case where the signal output is changing rapidly but smoothly.

Next, a case may be considered in which the information about the plosive sound becomes 1 (indicating that the plosive sound exists). This is also described in connection with FIG. When a plosive is present, a plosive adder (plosive adder) 403 arranges the output of the unsampled log representation within the time interval in which the plosive is known to be present. A fixed value is added to one or more specific samples of. For example, for a log representation of the signal output, it may be useful to use a fixed value of 1.2, and this value may be added to the signal of the log representation output at a time interval of 5 ms. May be. Figure 8
Exemplifies that a plosive sound is added in the case of a contour that is originally linearly interpolated. FIG. 9 exemplifies that a plosive sound is added in the case of a contour for each step. For the latter case, it is useful that the plosive be added after the step. Otherwise you will not be able to hear.

The exemplary embodiment of the invention described above includes two related but distinguishable classification processes. For example, as shown in FIG. 4, an output step evaluator (power step evaluator) 4
01 determines whether the output contour of the log representation between two consecutive samples can be linearly interpolated or a stepped contour can be presented. In addition, a plosive adder (plosive adder) 403 determines whether a plosive can be added to the output contour of the log representation between two consecutive samples. In other exemplary embodiments of the invention, any one of these processes may be performed independently of the other processes.

For clarity of explanation, the illustrative embodiments of the present invention are presented as including individual functional blocks or processors. The functionality presented by these blocks may be provided through the use of either shared or specialized hardware, including, but not limited to, hardware capable of executing software. For example, the functionality of the processors shown in FIGS. 1-4 may be provided by a single shared processor. (The use of the term processor is not to be construed as referring exclusively to hardware capable of executing software.)

An exemplary embodiment is the AT & T DSP16.
Alternatively, a digital signal processor (DSP), such as the DSP32C, read only memory (ROM) for storing software to perform the operations discussed above, random access memory (RAM) for storing results by the DSP. Can be included. Very large scale integrated circuit (VLSI) hardware embodiments may also be provided, as may customized VLSI circuits in combination with DSP circuits for general use.

While some of the specific embodiments of the present invention have been shown and described herein, these embodiments are merely the numerous possible specific embodiments that may be devised in application of the principles of the invention. It should be understood that this is merely an example of the arrangement of In accordance with these principles, numerous and varied other arrangements can be devised by those skilled in the art without departing from the spirit and scope of the invention.

According to the present invention, perceptually important characteristics can be transmitted for parameters in speech coding, and a more detailed description of parameter contours can be obtained. As a result, even when the signal output is transmitted at a relatively low bit transmission rate (bit rate), the quality of the signal to be reconstructed after being encoded is improved.

[Brief description of drawings]

FIG. 1 shows a general view of the transmission portion of an exemplary coding system with signal output as an explicit parameter and encoding according to an exemplary embodiment of the present invention.

FIG. 2 shows a general view of a receiving portion of an exemplary coding system having a signal output as an explicit parameter and encoding according to an exemplary embodiment of the present invention.

3 illustrates an exemplary plosive detector for use in the exemplary transmitter of FIG.

FIG. 4 illustrates an exemplary output (power) envelope processor for use in the exemplary receiver of FIG.

5 shows a so-called hat-hanging mechanism of the exemplary plosive detector of FIG. 3 operating in the absence of plosives.

6 is operated in the presence of a plosive, FIG.
2 shows a so-called hat-hanging mechanism of the exemplary plosive detector of FIG.

FIG. 7 shows a contour of a log-displayed signal output obtained by linear interpolation according to an exemplary embodiment of the present invention.

FIG. 8 shows the log output signal output contours and added plosives obtained by linear interpolation according to an exemplary embodiment of the present invention.

FIG. 9 shows the contour of the signal output in log representation obtained by stepwise interpolation, according to an exemplary embodiment of the invention.

FIG. 10 illustrates a log output signal output contour and added plosives obtained by stepwise interpolation in accordance with an exemplary embodiment of the present invention.

[Explanation of symbols]

101 Encoding Unit (Encoding Unit) 102 Output Extractor 103 Normalizer (Normalizer) 104 Encoding Unit (Encoding Unit) 105 Buzzer Detector 106 Low-pass Filter 107 Downsampler 108 Output Encoder (Output Encoder) 201 Decoder (Decoder) Unit 202 Output Extractor 203 Decoder (Decoder) Unit 204 Output Decoder (Power Decoder) 205 Output (Power) Envelope Processor 206 Scale Factor (Conversion Factor) Processor 207 Amplifier 301 Hat Hanger 302 Hat Keeper 303 Hat clearance detector 304 Peak clearance detector 305 Arithmetic unit 306 having logical AND function Down sampler having logical OR function 401 Output step evaluator (Power Step evaluator Eta) 402 upsampler 403 plosive adder (pop adder)

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Willem Bastian Crane               United States, 07920 New Jersey               Gee, Basking Ridge, Villaged               Live 87                (56) References JP-A-60-195600 (JP, A)                 JP-A-51-149706 (JP, A)                 JP-A-56-80099 (JP, A)                 JP-A-59-102297 (JP, A)                 JP-A 1-219895 (JP, A)                 JP-A-51-149706 (JP, A)                 Japanese Patent Publication Sho 62-39758 (JP, B2)

Claims (28)

(57) [Claims] 1. A method for decoding an encoded signal, wherein the encoded signal comprises an array of consecutive encoded parameter value signals indicative of a value of a predetermined parameter, Classifying the predetermined parameter into one of a plurality of categories based on a signal of the values of two consecutive coded parameters of the array; and based on the classified categories, Generating one or more intermediate parameter value signals indicating the value of the predetermined parameter one or more times between two consecutive encoded parameter value signals And the step of performing, the category includes a linear interpolation category and a step function category, and the predetermined parameter is divided into the linear interpolation category. Generating a signal of the intermediate parameter value, when digitized, numerically among the predetermined parameter values indicated by the two consecutive encoded parameter value signals. , Generating a signal of an intermediate parameter value exhibiting a value less than the greater one and greater than the lesser, wherein the predetermined parameter is classified into the category of the step function, The step of generating a signal of the intermediate parameter value has a value numerically equal to one of the values of the predetermined parameter indicated by the two consecutive encoded parameter value signals. A method of decoding an encoded signal, the method comprising: generating a signal of an intermediate parameter value. 2. The method of claim 1, wherein the encoded signal comprises an encoded speech signal. 3. Method according to claim 2, characterized in that the predetermined parameter represents the output of the audio signal. 4. The method of claim 3, wherein the predetermined parameter represents the output of a characteristic waveform. 5. The method of claim 1, wherein the step of classifying the predetermined parameter is the numerical difference between the values indicated by the signals of the values of the two consecutive encoded parameters. Classifying the predetermined parameter based on a method of decoding an encoded signal. 6. The method of claim 1, wherein the step of generating a signal of the value of the intermediate parameter when the predetermined parameter is classified into the category of the step function comprises a first intermediate value. Generating at least two intermediate parameter value signals, including a common parameter value signal and a second intermediate parameter value signal, the first intermediate parameter value And the signal of the value of the second intermediate parameter represent different numerical values for the predetermined parameter. 7. The method of claim 6, wherein the encoded signal comprises an encoded speech signal and the predetermined parameter represents an output of a characteristic waveform. A method of decoding a digitized signal. 8. The method of claim 1, wherein the encoded signal further comprises one or more of said one or more times between said two consecutive encoded parameter value signals. Including a coded parameter feature signal representing a value of the predetermined parameter, wherein the step of classifying classifies the predetermined parameter based on the coded parameter feature signal; A method of decoding an encoded signal, comprising: 9. The method of claim 8, wherein the encoded signal comprises an encoded speech signal. 10. A method according to claim 9, wherein the predetermined parameter represents the output of the audio signal. 11. The method of claim 10, wherein the plurality of categories represent the presence of audio signal output plosives and the absence of audio signal output plosives. A method of decoding an encoded signal, characterized in that it comprises a category. 12. A method of encoding a signal, the method comprising the steps of: generating an array of consecutive encoded parameter value signals indicative of a value of a predetermined parameter, The predetermined parameter is divided into a plurality of categories based on one or more times of the predetermined parameter values between two consecutive encoded parameter value signals. , Further, generating a signal characteristic of the encoded parameter based on the classified category, the plurality of categories, the plosive sound of the output of the audio signal A method of encoding a signal, characterized in that it comprises a category representing the presence and a category representing the absence of plosives in the output of the audio signal. 13. The method of claim 12, wherein the signal comprises an audio signal. 14. Method according to claim 13, characterized in that the predetermined parameter represents the output of the audio signal. 15. A decoder for decoding an encoded signal, wherein the encoded signal comprises an array of consecutive encoded parameter value signals indicative of predetermined parameter values. And the decoder classifies the predetermined parameter into one of a plurality of categories based on the following means: a signal of the values of two consecutive coded parameters of the array. Means for indicating the value of the predetermined parameter one or more times between the signals of the values of the two consecutive coded parameters based on the classified categories, Means for generating a signal of one or more intermediate parameter values, the categories including linear interpolation categories and step function categories, Means for generating said intermediate parameter value signal when said constant parameter is classified into said linear interpolation category, by said two consecutive encoded parameter value signals. Among the values of the predetermined parameter shown, including means for generating a signal of an intermediate parameter value that is numerically smaller than the larger one and greater than the smaller one, the predetermined one Means for generating the intermediate parameter value signal when the parameters of the are classified into the step function category are indicated by the two consecutive encoded parameter value signals. Means for generating an intermediate parameter value signal exhibiting a value that is numerically equal to one of the predetermined parameter values That, the decoder for decoding the encoded signal. 16. Decoder for decoding a coded signal according to claim 15, characterized in that the coded signal comprises a coded speech signal. 17. Decoder for decoding a coded signal according to claim 16, characterized in that said predetermined parameter represents the output of the audio signal. 18. Decoder for decoding a coded signal according to claim 17, characterized in that said predetermined parameter represents the output of a characteristic waveform. 19. The decoder of claim 15, wherein the means for classifying the predetermined parameter is numerically between the values indicated by the two encoded parameter value signals in succession. Decoder for decoding an encoded signal, characterized in that it comprises means for classifying said predetermined parameter based on a difference. 20. The decoder of claim 15, wherein the means for generating a signal of the value of the intermediate parameter when the predetermined parameter is classified into the category of the step function comprises: Means for generating at least two intermediate parameter value signals, including one intermediate parameter value signal and a second intermediate parameter value signal, said first intermediate A coded signal, wherein the typical parameter value signal and the second intermediate parameter value signal represent different numerical values for the predetermined parameter. Decoder for doing. 21. The decoder of claim 20, wherein the coded signal comprises a coded voice signal and the predetermined parameter represents an output of a characteristic waveform. A decoder for decoding the encoded signal. 22. The decoder of claim 15, wherein the encoded signal further comprises one or more times between the two consecutive encoded parameter value signals. Means for classifying the predetermined parameter, the coded parameter characteristic signal being representative of a value of the predetermined parameter, the means for classifying the predetermined parameter being based on the coded parameter characteristic signal; Decoder for decoding an encoded signal, characterized in that it comprises means for classifying the parameters. 23. Decoder for decoding a coded signal according to claim 22, characterized in that the coded signal comprises a coded speech signal. 24. Decoder for decoding a coded signal according to claim 23, characterized in that said predetermined parameter represents the output of a speech signal. 25. The decoder according to claim 24, wherein the plurality of categories represent a category indicating that there is a plosive sound at the output of an audio signal and a category indicating that there is no plosive sound at the output of an audio signal. Decoder for decoding an encoded signal, characterized in that it comprises a certain category. 26. An encoder for encoding a speech signal, the encoder means for: producing an array of signals of consecutive encoded parameter values indicative of the value of a given parameter. Means, between the two consecutive encoded parameter value signals in the array, one or more times, based on one or more of the predetermined parameter values; Means for classifying the parameter of one of a plurality of categories, further comprising means for generating a signal characteristic of the encoded parameter based on the classified category, Multiple categories, including a category that indicates the presence of an audio signal output plosive and a category that indicates an absence of the audio signal output plosive It characterized the door, encoder for encoding the signal. 27. The encoder of claim 26, wherein the signal comprises an audio signal. 28. The encoder of claim 27, wherein the predetermined parameter represents the output of the audio signal.